Most advanced process control systems are based on Model Predictive Control (MPC). In this paper we discuss three critical issues for the practical implementation of linear MPC for process control applications. The rst issue is related to oset free control and disturbance models; the second issue is related to the use of soft output constraints in MPC; and the third issue is related to the computationally ecient solution of the quadratic program in the dynamic regulator of the MPC. We have implemented MPC in .Net using C# and the MPCMath library. The implemented MPC is based on the target-regulator structure. It enables oset free control; it can be computed eciently on-line using several optimization algorithms; and accommodates soft constraint for the outputs and for shaping the set-point tracking penalty function. We report selected observations using this implementation and discuss their practical implications for process control. If the control and evaluation intervals are chosen too short, the predicted behaviour of the controllers may have unstable characteristics. Depending of the degrees of freedom, oset-free control of a number of the controlled variables can be achieved by introduction of noise models and integration of the innovation errors. If the disturbances increases, oset-free control cannot be achieved without violation of process constraints. A target calculation function is used to calculate the optimal achievable target for the process. The use of soft constraints for process output constraints in the MPC controllers, ensures feasible solutions. The computational load as function of controllers type, model dimension and constraint type are shown.
Proceedings of the 18th Nordic Process Control Workshop (npcw18), 2013